B[CITIC Construction Investment: Meta Muse accelerates the commercialization of intelligent agents, and the Computing Conference strengthens the full-stack AI layout] On September 29, CITIC Construction Investment’s research report pointed out that Meta Muse and the Computing Conference jointly strengthened the logic of AI application implementation to drive computing power demand. Meta launched Muse, a personal AI assistant that supports email, travel reservations and transaction execution. It was downloaded 2.8 million times in the first 12 days after it was launched. Meta also promoted integration with the Shopify payment ecosystem to promote the commercialization of consumer-grade intelligence. At the Yunqi Conference, Alibaba revealed that Qwen 4 is being trained, and the subsequent model is planned to evolve to 5 trillion-10 trillion parameters; it released the Zhenwu V900 chip, proposed a goal of global data center capacity exceeding 20GW in 2032, and promoted the implementation of enterprise intelligence through Qianwen Office. We believe that the upgrade of AI from dialogue to task execution is expected to continue to increase the demand for inference, tool invocation and cloud resources, focusing on high-speed optical modules, switches, servers, liquid cooling and power supplies. B[Alibaba updates flagship model Qwen3.8-Max] "Science and Technology Innovation Board Daily" reported on the 2nd that on September 2, Alibaba updated its flagship model Qwen3.8-Max, and its performance was significantly improved compared to the older version. After special post-training for programming (Coding) and professional office (Cowork), the overall performance of the new version of Qwen3.8-Max has been enhanced. In CodeArena, the world's authoritative three-party list focusing on front-end programming capabilities (WebDev), it increased by 22 points to 1691 points, ahead of Claude Opus5, Kimi K3 and other models, ranking first in the overall list. At the same time, CodeArena’s updated model price/performance list (Pareto frontier) shows that the new version of Qwen3.8-Max has a comprehensive average of only US$5 per million Tokens, directly "killing" all other models with prices greater than US$5. It is understood that Qwen3.8-Max is Alibaba Qianwen's most powerful large language model to date, with a total parameter of 2.4 trillion and support for 1 million context tokens. The updated Qwen3.8-Max has stronger agent programming capabilities and is more suitable for real complex enterprise tasks, scientific research, long-term tasks, etc.
— All loaded —
